Delphi: A Neuro-Symbolic Framework for Individualized, Safe and Interpretable Treatment Recommendation
Proceedings of the AAAI Conference on Artificial Intelligence · 14.03.2026 · 40 (23) · 19389-19397
DOI: 10.1609/aaai.v40i23.39016 ↗Kurz gefasst
Delphi-Studie in der Zeitschrift Proceedings of the AAAI Conference on Artificial Intelligence, online erschienen am 14.03.2026. Verfasst von Muchan Tao und 4 weiteren Autor:innen. Die Arbeit wurde bisher 3-mal zitiert (Crossref).
Abstract (Originalsprache)
Clinical reinforcement learning (RL) holds promise for treatment recommendation but remains hindered by black-box decision processes, limited safety guarantees, and lack of individualized reasoning. We introduce Delphi Engine, the first fully trainable neuro-symbolic causal RL framework for dynamic treatment planning, designed to answer three core clinical questions in real time: Why this action? Why is it safe? Why for this patient? Specifically, Delphi integrates: (1) causality-aware state modeling using discretized physiological variables and subtype-specific causal graphs; (2) adaptive symbolic rule constraints, combining clinical guidelines and behavior-derived rules into soft differentiable logic; and (3) interpretable decision fusion, where actions are selected based on joint neural-symbolic Q-values and explained via structured LLM-based justifications. We evaluate Delphi on the MIMIC-III sepsis cohort using both standard off-policy evaluations (WIS↑1.47, DR↑1.29, RMSE↓0.207) and the first blinded physician evaluation of an explainable RL system in healthcare. Delphi consistently outperforms historical physicians' treatments in safety (+10.4%), understandability (+8.9%), and adoption rate (+5.75%) across six clinical axes. These results highlight Delphi’s potential as a safe, interpretable, and patient-specific AI assistant for critical care medicine.
Zitieren
APA
Tao, M., Qin, H., Fang, Y., Shan, C., & Tan, T. (2026). Delphi: A Neuro-Symbolic Framework for Individualized, Safe and Interpretable Treatment Recommendation. Proceedings of the AAAI Conference on Artificial Intelligence, 40(23), 19389-19397. https://doi.org/10.1609/aaai.v40i23.39016
Vancouver
Tao M, Qin H, Fang Y, Shan C, Tan T. Delphi: A Neuro-Symbolic Framework for Individualized, Safe and Interpretable Treatment Recommendation. Proceedings of the AAAI Conference on Artificial Intelligence. 2026;40(23):19389-19397. doi:10.1609/aaai.v40i23.39016
BibTeX
@article{Tao2026,
author = {Tao, Muchan and Qin, Haonan and Fang, Yuqi and Shan, Caifeng and Tan, Tieniu},
title = {Delphi: A Neuro-Symbolic Framework for Individualized, Safe and Interpretable Treatment Recommendation},
journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
year = {2026},
volume = {40},
number = {23},
pages = {19389--19397},
doi = {10.1609/aaai.v40i23.39016}
}
EUVIMED